loading...

re: Lost potential of personal computing VIEW POST

TOP OF THREAD FULL DISCUSSION
re: Very good observation! I think a large part of the problem is that we no longer care about optimization. Too often, we do less in a program that t...
 

As someone who started computing with C64, the optimization issue is very true, and sometimes drives me crazy. Also, we have way too more abstractions, so doing simple things are not very easy anymore. This abstraction is result of software reuse, collaborative mind and commercialization which needs speed (more job done in less time) and results.

Because of a single sentence ("Hardware is cheap, people are expensive"), I refused to learn Python for years. I still write all my applications in C++. Only toys are written in Python and ported to C++ after they matured.

With this latest Electron craze, we are running whole browser instances to highlight text in various colors. Atom starts up with 660MBs of RAM. Only Eclipse is topping it with 720MBs, which is a proper IDE, and is acceptable in my book. When written correctly, a 2MB application with 170MB RAM usage can make very hard and accurate scientific calculations while kneeling down whole CPUs and using them effectively.

BTW, PalmOS and Palm devices are very very powerful for their era. I used a Handspring Neo and PalmOne LifeDrive for a very long time.

I will leave two comics here. One is an "obligatory XKCD".
Obligatory XKCD!
Just running slack

 

Ha! Very true.

Now, with all that said, I am in fact a Python developer as well. Performance isn't entirely lost in that language, especially if you're running the Pypy interpreter (instead of the standard Python interpreter, which is much slower). There are many abstractions the language does well, including immutable objects, which make for serious performance advantages. The thing I like about Python is that you can come to understand what's going on under the hood, and make wise performance decisions as a result.

In my opinion, C, C++, and Python all have their places in responsible development, and I wouldn't regard any of them as only good for "toys". You just need to know when to use which, and how to get the most out of them.

Of course, horses for courses. I'm working on scientific, high performance computing both as a researcher and system administrator.

Python has improved a lot in ML, DL areas and very performant with underlying platform native libraries and PyPy, you're true on that regard. Also it's joyful to write Python, because it really does some tasks well with few lines and with great felxibility. It's a very productive language.

However, I generally work on very big matrices filled with double precision floating point numbers and perform operations on these, and C++ is much faster in these cases, at least in my domain.

My relationship with Python is much better now, but if I'm coding something serious I'd use Python for prototyping. If it's a small utility or something, I'd directly implement it in Python. However I'm very resource concious while developing applications, so I'm still not very comfortable to leave an interpreted language 7/24 as a service at the background.

Also, I'd like acknowledge that modern C++ isn't the easiest tool to work with. It's like a double edge blade with no handle. :D

P.S.: Yes, Calling Python a language for "programming toys" was a wrong and childish. Maybe I should be clearer on that part, sorry.

Edit: Writing comments fast while being tired is not a good idea. Corrected all the half sentences, polished the structure. IOW refactored the post completely.

Great clarification. I think we're very much on the same page regarding C++ and Python, and their respective uses. I've been the lead on a game engine project for a while, and we use C++ for that, not Python. We need the additional control.

However, GUI is almost always a lot friendlier in Python than in C++, so I tend towards that for standalone application development.

Incidentally, C++14 and C++17 have added a number of "handles" to that double-edged blade. For example, std::unique_ptr and std::shared_ptr handle a lot of memory safety (and deallocation) challenges quite deftly.

P.S. You'd probably find some stuff interesting in PawLIB, especially the development version (v1.1). It attempts to merge performance with some sweeter syntax.

Thanks, it seems so. GUI was never a strong part of C++ programming, but I intend to learn Qt both on Python and C++. I've some small projects in my mind. Also, I really want to be able to use ncurses on C++.

I have big C++11 code base I'm working on (sorry, can't give many details about it but it's a scientific code), and I may migrate to the new pointers, but I must do extensive testing on performance. Some parts of the code is very very sensitive and very hot, so fiddling with these parts affects just about everything.

Thanks for the pointer to the PawLIB. It really looks interesting, however it doesn't address any of my bottlenecks for now. If I find anything limiting, will give its data structures a try. I'm using Eigen for matrices and algebra, Catch for testing, and Easylogging++ for logging. They addressed what I need so far.

code of conduct - report abuse